Abstract

The enhanced Fujita scale category 4 (EF4) Tuscaloosa, Alabama, tornado on 27 April 2011 produced 64 fatalities along its 130-km track. Hybrid survey/interviews were conducted with a sample of 211 Tuscaloosa-area residents to determine how the 27 April tornado might change future shelter-seeking plans. Despite a history of tornadoes in the area, only 47% of Tuscaloosa residents had shelter plans in place prior to 27 April, but 62% intend to change their shelter plans or have shelters plans for the future. Changes in shelter-seeking plans were divided into four groups and discussed according to commonalities. Logistic regression with demographic variables was then used to predict those likely to have shelter plans before 27 April and those likely to change their shelter plans in the future. Among these variables, residents over age 55 [odds ratio (OR) 8.9, 95%; confidence interval (CI): 2.167–36.352] and those having a bachelor's degree (OR 5.1, CI: 1.342–19.316) were more likely to have had shelter plans before 27 April. The most significant variable indicating a change in future shelter-seeking plans is being Hispanic/Latino (OR 5.2, CI: 1.753–15.465). These results may assist National Weather Service (NWS) personnel, broadcast meteorologists, emergency managers, and city planners with the development of targeted warning communication tactics and safety strategies for a future tornado event.

1. Introduction

Spring 2011 was defined by the anomalous hyperactivity of tornadic thunderstorms in the United States. For the first five months of 2011, there were 1217 confirmed tornadoes and 539 fatalities nationwide, making 2011 the deadliest tornado year since 1953 (Storm Prediction Center 2011a). The Mississippi/Alabama Super Outbreak on 27 April 2011 was responsible for 226 confirmed tornadoes, 339 fatalities, and thousands of injuries (Storm Prediction Center 2011b). An unusual aspect of the spring 2011 tornado season was the number of tornadoes that tracked into larger metropolitan areas. As most southeastern and Great Plains cities continue to sprawl, the probability of densely populated areas being impacted by tornadoes is increasing over time (Hall and Ashley 2008).

Such tragedy has spawned renewed emphasis on how people perceive, respond, and react to the threats posed by tornadoes and other severe weather hazards. In particular, warning communication (Sorenson 2000; Hammer and Schmidlin 2002; Brotzge and Erickson 2009; Coleman et al. 2011; League et al. 2010; Schumacher et al. 2010; Sherman-Morris 2010; Hoekstra et al. 2011), perception (Senkbeil et al. 2010; Silver and Conrad 2010), and false alarms (Dow and Cutter 1998; Barnes et al. 2007a,b; Simmons and Sutter 2009; Brotzge et al. 2011) are active topics of mutual research. Additionally, Schultz et al. (2010) used hypothetical scenarios to determine how people might respond during an actual tornado event. Respondents were presented with a tornado-at-home scenario and a tornado-while-driving scenario. For the at-home scenario, most respondents (82%) preferred to stay in the home, and for the driving scenario most respondents (72%) indicated that they would leave their vehicles to seek shelter.

No previous research has been directly devoted to changes in individual shelter-seeking plans of residents after a significant tornado event. Using Tuscaloosa, Alabama, as a case study, the intent of this research is to illustrate the specific changes to shelter-seeking plans that residents may implement for a future tornado event. Thus, the crux of this research evaluates possible behavioral changes of survivors; whereas, much previous research has considered reasons for fatalities. Nevertheless, it is important to also understand factors contributing to fatalities in previous research, since that is also related to possible behavioral changes of survivors.

Biddle (2007) is perhaps the most comprehensive evaluation of situational and cognitive variables leading to tornado fatalities. Among situational risk factors, residents of lower socioeconomic status, residents living in manufactured homes, residents living near the start of the tornado path, and elderly or disabled were associated with higher fatalities. Schmidlin and King (1995) analyzed risk factors that contributed to fatalities during the March 1994 tornadoes in Alabama and Georgia with many similar conclusions. Fatalities were greater for residents that were older, living in mobile homes, watching television less, and aware of the approaching tornado for less time compared to residents who survived. Poor building anchorage and not being located in a basement contributed to fatalities during the 31 May 1985 outbreak (Carter et al. 1989). Glass et al. (1980) was one of the first studies to assess factors contributing to fatalities after the 1979 Wichita Falls, Texas, tornado. The risk of serious injury or death was higher for residents who attempted to drive out of the storm's path and also for mobile home residents. While there is little disagreement regarding the conclusions about the hazards faced by residents of mobile homes (Schmidlin et al. 2009; Chaney and Weaver 2010; Sutter and Simmons 2010), the meteorological community is conflicted on the effectiveness of using a vehicle as a place of last refuge (Schmidlin et al. 1998, 2002). With ample warning lead time, a vehicle is a reasonably safe place of refuge. This is true, provided that the driver has knowledge of the tornado warning polygon or that the driver is listening to local meteorologists describing landmarks and times of the tornado's forecasted location.

Tornado warning communication and response, tornado preparedness, and tornado hazard perception are all important interrelated topics that help explain the protective action decision making of residents (Lindell and Perry 2004). The personal decision to seek shelter is made in the minutes after a tornado warning is issued. There have only been a few studies that concentrate on shelter-seeking behavior. Hammer and Schmidlin (2002) investigated the locations and fatality distribution of where individuals sought shelter for the 3 May 1999 Oklahoma City, Oklahoma, tornado. Schmidlin et al. (2009) explored tornado shelter-seeking options and shelter-seeking behavior of mobile home residents. Balluz et al. (2000) analyzed shelter-seeking actions for the 1 March 1997 Arkansas tornadoes. Four factors were positively associated with shelter-seeking actions: 1) education level, 2) having a basement, 3) hearing a warning siren, and 4) having a household response plan. Similarly, Blanchard-Boehm and Cook (2004) identified significant predictor variables of protective measures for future tornadoes. None of these studies explicitly discuss how residents might change their future plans if faced with another tornado event.

This research has two specific objectives. The first objective is to use the results from interview responses to track how individuals might change their shelter plans for a future tornado event. The second objective is to use the results from demographic survey questions with logistic regression to predict certain characteristics of residents that are likely to change their shelter plans for a future tornado event. With a combination of qualitative individualized tracking and statistical results, it is hoped that National Weather Service (NWS) personnel, broadcast meteorologists, emergency managers, and city planners may use this research to make decisions about strategies to target particular groups of residents in warning statements.

2. Background and tornado history of Tuscaloosa County

Tuscaloosa, Alabama, was the largest city to be directly impacted by a violent tornado on 27 April 2011. Together with its river companion, Northport, Alabama, the metropolitan population of Tuscaloosa is estimated to be approximately 150 000 (Fig. 1). This number includes the University of Alabama, which adds 30 000 students to the population between August and May. Although the area is frequently impacted by tornadoes, the downtown business district of Tuscaloosa and the University of Alabama campus had not been threatened since 1932. The intensity of the tornado progressed from the enhanced Fujita scale category 3 (EF3) as the tornado entered the county from the west to upper EF4 (190 mph) as it exited the city limits of Tuscaloosa (National Weather Service Birmingham 2011). The tornado heavily damaged or destroyed large residential sections of the Rosedale, Forest Lake, Alberta City, and Holt communities within Tuscaloosa. Many of these communities were home to a high proportion of renters and residents of lower socioeconomic status.

Fig. 1.

Tuscaloosa County population by ZIP code with the 27 Apr 2011 tornado swath.

Fig. 1.

Tuscaloosa County population by ZIP code with the 27 Apr 2011 tornado swath.

Since 1950 there have been 7 tornadoes within the city of Tuscaloosa and 38 within Tuscaloosa County (Fig. 2). The spatial distribution of tornado tracks in Tuscaloosa County shows two regions of higher density: the northwestern side of the county extending into the city of Northport and the south side of Tuscaloosa. The south side of Tuscaloosa was impacted by an Fujita scale category 4 (F4) tornado in 2000, which generated a large number of applications for residential storm shelters and safe rooms (D. Hartin, Tuscaloosa County Emergency Management Agency, 2011, personal communication).

Fig. 2.

Spatial distribution of tornado tracks within Tuscaloosa County 1950–2009. The 27 Apr 2011 tornado is indicated by the black swath, and other significant historical tornadoes are labeled.

Fig. 2.

Spatial distribution of tornado tracks within Tuscaloosa County 1950–2009. The 27 Apr 2011 tornado is indicated by the black swath, and other significant historical tornadoes are labeled.

During our surveys/interviews, numerous respondents revealed a perception of two tornado alleys within the county that correspond with the spatial distribution in Fig. 2. This information was volunteered by numerous residents during interviews when many respondents wanted to tell their stories. Thus, we began to note and discuss this recurring theme related to previous tornado tracks in the area; albeit, we did not document the total number of times this particular comment was heard by each researcher. Although we did not directly ask about previous tornado experiences, respondents were asked to indicate if they had ever been under a tornado warning but did not experience a tornado. A total of 90% of respondents indicated this was the case, and this is not surprising given current false-alarm ratio statistics (Brotzge et al. 2011), and the amount of activity in Tuscaloosa and the surrounding counties between 2000 and 2011. There were numerous spring 2011 squall-line events with weak tornadoes, and a significant tornado outbreak on 15 April 2011. A long-track 15 April 2011 tornado that weakened to EF1 intensity in the city limits of Tuscaloosa followed a track almost identical to the 16 December 2000 F4 tornado. Unfortunately, prior to the 27 April 2011 tornado, there appeared to be a pervasive belief that tornadoes always travel south or north of the city center of Tuscaloosa based on recent and historical events. This belief was shared by Tuscaloosa residents from a variety of demographic categories in our interviews.

3. Methods

a. Survey/interview development and implementation

A 22-question hybrid survey–interview was developed with 5 open-ended questions and 17 closed questions. Since the survey is too lengthy to easily make into a figure, an abbreviated version with only the questions used for this article is presented (Table 1). The first 7 questions of the survey ask about demographic information. The remaining questions ask about shelter plans and severity of home damage on 27 April. To improve the efficiency of the surveying process, the survey was converted to an hypertext markup language (HTML) platform using Zoomerang (www.zoomerang.com) and was offered to respondents via iPads. Respondents could easily navigate between questions with input from interviewers, or interviewers would read questions while allowing respondents to view the questions and choices. All responses were stored on the Zoomerang server. Informal conversation regarding shelter adequacy and future strategy was often discussed upon completion of interviews. Respondents were informed about residential and community safe rooms and referred to the Federal Emergency Management Agency (FEMA) Building Sciences website for additional information (FEMA 2012). This type of interaction was beneficial for both the respondent and the researcher.

Table 1.

Closed and open-ended questions used during the survey and interview process.

Closed and open-ended questions used during the survey and interview process.
Closed and open-ended questions used during the survey and interview process.

The days after the tornado were chaotic as Tuscaloosa began to mobilize toward recovery and debris removal efforts. Several distribution points were established throughout the city. At these distribution points, teams of volunteers from various organizations assembled supplies and then drove out into the affected neighborhoods to find people in need. The city also opened numerous shelters for those rendered homeless. Our team of eight researchers decided to adopt a strategy of targeting shelters and distribution points to find residents directly affected by the tornado. For a 2-week period in early May 2011, half of the team rode into neighborhoods accompanying distribution teams, while the other half concentrated on residents in shelters. Although the data collected represent a convenience sample, this strategy provided our team with the best opportunity to connect with residents impacted by the tornado. Surveys/interviews with less-affected residents also occurred at several locations around the Tuscaloosa area for the same 2-week period in early May.

A total of 211 completed surveys/interviews were collected and used in this research. Demographic characteristics of respondents approximate the 2010 census composition of Tuscaloosa with a few exceptions (Fig. 3). The counts reported in Fig. 3 represent the conversion of the 2010 census population percentages for the city of Tuscaloosa applied to the sample of 211 respondents. For example, the population of the city of Tuscaloosa is 90 468 with 46 956 females (52%), which indicates that there should be approximately 110/211 females for a representative sample.

Fig. 3.

Demographic distribution of respondents for (top) education, (second from top) race/ethnicity, (third from top) age, and (bottom) gender compared to estimated values for the city of Tuscaloosa from the 2010 census.

Fig. 3.

Demographic distribution of respondents for (top) education, (second from top) race/ethnicity, (third from top) age, and (bottom) gender compared to estimated values for the city of Tuscaloosa from the 2010 census.

The Hispanic/Latino population appears to have been oversampled at first glance, but the census results likely did not accurately capture the growing Hispanic/Latino population of Tuscaloosa prior to the passage of the Hammon–Beason Alabama Taxpayer and Citizen Protection Act (Alabama House Bill 56), a controversial illegal immigration law. Every effort was made to increase the sample size of the African American population, but this group declined interviews more often than other groups. Additionally, our research team encountered few residents over age 55 at residences or in shelters, while 25–34-year-olds were numerous (Fig. 3).

b. Data analysis

Data analysis consisted of two approaches. Responses to open-ended questions were best summarized and discussed qualitatively. Logistic regression was also used for the purpose of predicting the likelihood of future shelter plans based on social and demographic variables. The combination of these two methods identifies specific individual approaches to changes in shelter plans while also providing an estimate of the likelihood that certain populations will change their plans.

Qualitative analysis involved responses to questions 11, 15, and 16 from Table 1. In question 11, if respondents answered that they had shelter plans prior to 27 April 27, the respondents were then asked to discuss their plans. These responses were then categorized into groups. The process was then iterated for responses to questions 15 and 16 from Table 1. Responses from question 16 were then directly compared to responses from question 11 for each individual respondent. This allowed for the tracking of how each individual might change her plan for a future event.

Any evaluation of existing shelter plans and changes in future plans depends on the definition of a shelter plan. Respondents were not provided a list of shelter plans in a survey question. It was an open-ended question, and responses were recorded and grouped by common answers. For example, if the respondent answered a car was his shelter, researchers determined that the car was used to drive out of the path and not used as a stationary shelter. There were no respondents identified as already being in their vehicles on 27 April prior to the approach of the tornado. Similarly, if a respondent answered that he had a shelter plan but discussed diving for cover inside the house at the last second, it was recorded as having no shelter plan.

Counts were summed for several categories based on the responses to sheltering plans. The first question had only two categories: the number of residents that had shelter plans before 27 April and the number of residents who indicated that they would change their shelter plans for a future event. Residents were then divided into four groups. These groups are 1) Yes, I had a shelter plan; No, I am not going to change it; 2) Yes, I had a shelter plan; Yes, I am going to change it; 3) No, I did not have a shelter plan; Yes, I am going to change it; and 4) No, I did not have a shelter plan; No, I am not going to change it. Counts were then summed for each category (Fig. 4). Next, responses in each category were grouped by shelter answers to facilitate analysis. For example, 13 residents in the (yes, yes) category plan to apply for a residential storm shelter or safe room in the future (Table 2).

Fig. 4.

Counts of respondents organized by groups of shelter plans before and after 27 Apr 2011. Groups include 1) No, I did not have a shelter plan; No, I am not going to change it; 2) No, I did not have a shelter plan; Yes, I am going to change it; 3) Yes, I had a shelter plan; No, I am not going to change it; and 4) Yes, I had a shelter plan; Yes I am not going to change it.

Fig. 4.

Counts of respondents organized by groups of shelter plans before and after 27 Apr 2011. Groups include 1) No, I did not have a shelter plan; No, I am not going to change it; 2) No, I did not have a shelter plan; Yes, I am going to change it; 3) Yes, I had a shelter plan; No, I am not going to change it; and 4) Yes, I had a shelter plan; Yes I am not going to change it.

Table 2.

Individual shelter-seeking change locations organized by shelter group. NA = not available; IDK = I do not know.

Individual shelter-seeking change locations organized by shelter group. NA = not available; IDK = I do not know.
Individual shelter-seeking change locations organized by shelter group. NA = not available; IDK = I do not know.

Logistic regression with demographic variables was used to predict the likelihood of having a shelter plan before 27 April and the likelihood of changing a shelter plan in the future. For both models, event-per-variable ratios exceeded the recommended minimum of 10 (Perduzzi et al. 1996). The regression equation for predicting who was likely to have a shelter plan contained all variables in a forced entry method using SPSS (formerly Statistical Package for the Social Sciences) version 19. These variables included race, age, gender, education, years lived in Tuscaloosa, marital status, and children living in the home, with shelter as the dependent variable. All variables were included in this first exploratory analysis. It was important to ascertain the interrelatedness of all social and demographic variables in predicting who was likely to have a shelter plan. Furthermore, elimination of insignificant variables slightly decreased the overall significance of the model.

Criteria were placed on variable inclusion for the analysis of who was likely to change their shelter plans. The variables in this analysis were the same as the first step with the addition of home damage as a predictor variable and change in shelter plan as the dependent. A forward-stepwise method was used with a Wald statistic and an alpha level of 0.05 for entry, and 0.10 for removal in SPSS version 19. The forward method eliminated all variables but race and age. This procedure was amended by adding home damage and marital status into the model. These variables were not significant; however, categories within the variables were nearly significant. The addition of home damage and marital status slightly improved the overall significance of the model.

4. Results and discussion

a. Possible changes to individual shelter plans

The hybrid survey/interview contained many open-ended questions. Residents were asked if they had shelter plans before 27 April and also to discuss those plans. They were then asked if they would change their plans for a future severe weather event when tornadoes are likely and to discuss future plans. Counts were summed for several groups based on the responses to sheltering plans.

Of the 211 residents surveyed, 99 (47%) had shelter plans before 27 April 27. This percentage is surprising given the common occurrence of tornado fatalities in the southeastern United States (Ashley 2007). Of the respondents who indicated having shelter plans before 27 April 2011, the vast majority of plans were closets and bathrooms of permanent homes. Exceptions were the small number of respondents having residential safe rooms and respondents using vehicles to drive away (Table 2). After 27 April, 131 (62%) of residents said that they would change their shelter plans for a future event. Residents were then divided into four groups for qualitative analysis as described in the methods section (Table 2). An additional cross tabulation of the four groups with counts within selected demographic variables provides greater detail of the composition of the four groups (Table 3).

Table 3.

Cross tabulation of counts for the four shelter groups and selected demographic variables.

Cross tabulation of counts for the four shelter groups and selected demographic variables.
Cross tabulation of counts for the four shelter groups and selected demographic variables.

In all groups, [I do not know (IDK)] was the most common response when asked to discuss changes to an existing shelter plan. This was especially true in the (no, yes) group with 58 IDK responses. Many of the respondents in this group indicated that they would do something different in the future, but they did not know what action to take at that time. In the (yes, no) group, 14 respondents had shelter plans but failed to provide details about their plans, resulting in [no answer (NA)]. This number is explained by a slow and intermittent wireless signal during the first day of field surveys/interviews. The disruption in wireless service on day 1 resulted in longer interview times with respondents only willing to answer shorter survey questions. After day 1, any technological iPad and Zoomerang issues were resolved enabling efficient data collection.

The IDK and NA answers were not included in the reporting of counts for each subcategory. There were 38 responses in the (yes, no) subcategory, and 22 responses were either the bathroom or an interior hallway or closet. A total of six residents had a storm shelter on their property (Table 2). Despite the visual evidence of hundreds of destroyed homes with failed interior rooms in Tuscaloosa, many residents plan to take their chances in an interior room if another violent tornado strikes the area in the future. The majority of fatalities in Alabama on 27 April 2011 were from people sheltering in permanent homes (Storm Prediction Center 2011a). Thus, most residents sought shelter in recommended interior rooms unaware of how inadequate that strategy can be on days when violent EF4 and EF5 tornadoes are expected. In the absence of a residential safe room or storm shelter, an interior room of a well-built permanent home should be the recommended shelter for typical severe weather days, but an alternative shelter is suggested for days with violent tornadoes.

In the (yes, yes) group, all the residents answered that their plans were an interior hallway, closet, or bathroom prior to 27 April. A total of 13/26 (yes, yes) residents said that they would either construct a residential storm shelter in the future or apply for a residential safe room. The most popular models in Alabama are external in-ground residential storm shelters (occupancy 6–8) normally built into the side of a small hill. Five of the residents plan to drive to a safe building or to drive out of the tornado path. On days when violent tornadoes are likely, all of the stated future shelter plans are an improvement from the existing plan of an interior room.

The (no, yes) group is perhaps the most important indicator of shelter plan change in the future. Unfortunately, this group had the highest number of IDK responses. Since the research was conducted so soon after 27 April 27, it is hoped that these residents have established shelter plans for the future if they did not know at that time. Many residents were mentally and physically exhausted during the recovery phase and were only willing to answer shorter survey questions. Of the 26 residents that responded in this group, 11 plan to use a vehicle to either drive out of the path to a community shelter or to a perceived sturdy building. Seven residents plan to apply for FEMA mitigation grants to construct residential safe rooms. The cost of a safe room varies for individuals and by safe room type and design. The recommended 8 ft × 8 ft safe room that also functions as a closet ranges in price from $6600 to $8700 (2011 U.S. dollars). Certain individuals meet requirements to have up to 75% of the total cost offset (FEMA 2012). Many impacted Tuscaloosa residents lack sufficient property size for an external shelter and lack sufficient income to install an internal safe room.

The (no, no) group displayed sentiments of fatalistic attitudes and hopelessness once thought to be a regional phenomenon (Sims and Bauman 1972). It should be noted that this was the smallest group, and that other regions would be likely to show similar results (Cohen and Nisbett 1998). Upon completion of surveys/interviews, willing respondents in the (no, no) group were engaged in conversations about possible shelter options and safer strategies for the future. A few were surprised to learn about aboveground safe rooms or other storm shelters capable of withstanding winds from violent tornadoes. These respondents mentioned that they might check into those options. Most implied or stated that life or death was in the hands of a divine power.

Sadly, some people indicating a change in shelter plan answered that they would go to big-box buildings, such as Lowes, Wal-Mart, or Home Depot, for a sturdy refuge area. Destruction of big-box buildings from tornadoes in 2011 (Sanford, North Carolina; Tuscaloosa, Alabama; and Joplin, Missouri) proves the fragility and danger of such structures in high winds. Basements were common responses for all three groups. Basements that are not completely underground on all sides or that have wooden floors overhead are vulnerable during a violent tornado. The increasing response of residential safe rooms and shelters is encouraging to see, provided that residents are choosing shelter models from manufacturers that have been determined safe by wind engineers.

It is also encouraging and discouraging to see that residents are considering driving out of the path of the storm. This strategy would have worked for most residents during the Tuscaloosa tornado and other tornadoes with long warning lead times (Hammer and Schmidlin 2002; Hoekstra et al. 2011). Caution should be used in employing this approach, so that residents can understand the necessary amount of time required to drive away. Furthermore, some people may misinterpret storm warning polygon information and possibly drive into harm's way. Additionally, considerable traffic congestion would result if 20 000–30 000 urban residents simultaneously decided to use vehicles to flee from a tornado. However, the risk of fatalities, severe injuries, and minor injuries was lower for those fleeing their homes in vehicles than those remaining in their homes during the 3 May 1999 Oklahoma tornado outbreak (Daley et al. 2005). This suggests that a reliable vehicle is a viable shelter option on days when violent EF4 and EF5 tornadoes are expected to destroy permanent homes in suburban and urban areas. Driving away is not a shelter plan that can be used with consistency by all residents in all situations, but it can be used effectively in certain situations, especially by residents of mobile homes. Because of its mobility and minimal structural protection, a vehicle is a more appealing shelter option than lying in a ditch, where other environmental hazards can be a concern (Schmidlin et al. 2009).

b. Predicting changes to shelter plans

Two separate logistic regression models were created using demographic and social variables to predict who was likely to have a shelter plan before the event and who was likely to change their shelter plans after the event. The first model attempted to predict who was likely to have a shelter plan before the event. All variables were forced into the model for this first analysis. The model was not a good predictor of who had a shelter plan prior to 27 April 2011, with only 69% of the respondents correctly classified. Numerous model diagnostics, such as −2 log-likelihood, Nagelkerke r2, and a Hosmer and Lemeshow test, revealed inadequate model fit. Furthermore, the 95% confidence interval (CI) on the odds ratios (ORs) for most variables spans the number 1, indicating the possibility of no change in odds.

Variables that contributed most to the likelihood of having a shelter plan before the event were age and education. Age was the only variable to be significant overall. Within the age and education variables, there were some significant categories. Within the age variable, 35–44-year-olds (OR 3.8, CI: 1.33–10.771) and residents over age 55 (OR 8.9, CI: 2.167–36.352) were more likely to have a shelter plan when compared to the reference category of 19–24-year-olds (Table 4). Although not significant, other age groups also displayed positive odds ratios. Males were less likely than females to have had a shelter plan, but this result was not significant at the 0.05 level. Similar results were reported in Sherman-Morris (2010).

Table 4.

Logistic regression results for residents likely to have had shelter plans prior to 27 April 2011. Significant values are in bold. Single asterisk means <.05; double asterisks mean <.01.

Logistic regression results for residents likely to have had shelter plans prior to 27 April 2011. Significant values are in bold. Single asterisk means <.05; double asterisks mean <.01.
Logistic regression results for residents likely to have had shelter plans prior to 27 April 2011. Significant values are in bold. Single asterisk means <.05; double asterisks mean <.01.

For the likelihood of having a shelter plan within the education variable, odds ratios increased as education increased, with the notable exception of residents with a graduate degree. The most significant category when compared to the category of high school or less was bachelor's degree (OR 5.1, CI: 1.342–19.216) followed by some college/associate's degree (OR 3.9, CI: 1.196–12.607).

The second model attempted to predict the likelihood of residents to change their shelter plans in the future. A forward entry/removal method was used in this step. Only the variables race and age met the criteria for inclusion. Although not significant overall, the variables home damage and marital status each contained a nearly significant category, so these two variables were added to improve the overall significance of the model. Eighty-seven percent of the shelter-changing respondents were accurately classified. The −2 log-likelihood score was lower (247.697), Nagelkerke r2 was higher (0.212), and a Hosmer and Lemeshow test revealed better model fit (X2 6.736, p = 0.565) than the model for shelter plan change with all variables included. Thus, the final model for change in shelter plan included race, age, home damage, and marital status. Even though the second model was a better fit, the 95% confidence intervals are broad because of the small sample sizes within each group. If our sample was larger, then the confidence intervals would likely become narrow, providing a more accurate representation of past and future predictors.

Although race was not a significant predictor of having a shelter plan before the event, Hispanic/Latino residents in particular appear to be more likely to change their shelter plans in the future when compared to the reference category of white residents (OR 5.2, CI: 1.753–15.465) (Table 4). The Alberta City section of Tuscaloosa (ZIP code 35404) (Fig. 5) was heavily populated by Latino residents, and the tornado was EF4 intensity there. Consequently, damage in that part of the city was more extensive. A separate article will address tornado hazard perception of Hispanic/Latino residents in Tuscaloosa.

Fig. 5.

Distribution of respondents for Tuscaloosa area ZIP codes.

Fig. 5.

Distribution of respondents for Tuscaloosa area ZIP codes.

Residents age 55 and older were less likely to change their shelter plans in the future (OR 0.26, CI: 0.068–0.974). This is presumably due to satisfaction with their existing shelter plans and the much greater likelihood of people in that demographic having a plan before the event. Residents whose homes were completely destroyed on 27 April were more likely to change their future shelter plans compared to residents with no home damage; however, this result was not significant at the 0.05 level. Curiously, residents who experienced some or significant damage to their homes were not likely to change their shelter plans.

An example of an odds ratio calculation is provided here for clarification. Using four predictors of possible changes in future shelter plans, the following equation estimates the probability of shelter plan change for a Hispanic/Latino resident, whose home was destroyed, age 35–44, and married. Values are taken from the coefficients listed in Table 5. Here,

 
formula

where Z = 0.182 (constant) + 1.650(1) + 0.915(1) + 0.511(1)–0.665(1).

Table 5.

Logistic regression results for residents likely to change their shelter plan in the future. Significant values are in bold. Single asterisk means <.05; double asterisks mean <.01.

Logistic regression results for residents likely to change their shelter plan in the future. Significant values are in bold. Single asterisk means <.05; double asterisks mean <.01.
Logistic regression results for residents likely to change their shelter plan in the future. Significant values are in bold. Single asterisk means <.05; double asterisks mean <.01.

The probability of shelter plan change is 0.93.

A calculation where the Hispanic/Latino term is substituted with the African American term with all other variables remaining the same reveals the following changes in the odds ratio:

 
formula

The odds from substituting African American for Hispanic/Latino and holding all other variables the same is

 
formula
 
formula
 
formula

Hispanic/Latino residents are 4.2 times more likely to change their shelter plans when compared to African Americans in the same demographic categories.

c. Limitations

There are some limitations to the results of this research. Respondents could have introduced bias by providing answers to sheltering questions that were met with approval by researchers. Expressed and revealed preferences suggest that what respondents say they will do in the future is unreliable (Tobin and Montz 1997). These concerns are valid. All researchers asked questions and recorded responses with a uniform approach, providing only clarification about the meaning of the question to respondents so as not to introduce bias. It is hoped that our consistent approach mitigated some of these concerns.

A stated change in a future shelter plan may not translate into an actual behavioral change in shelter plan. The best way to assess behavioral change is through additional work with the same respondents. One of the time-sensitive goals of this research was to survey and interview people in the weeks following the event so that their actions on 27 April were more accurately recalled. A rapid Institutional Review Board (IRB) approval was expedited by eliminating any questions that would possibly allow respondents to be identified, hence also eliminating the potential for future contact with the same respondents. Another potential limitation was knowing whether people sought shelter alone or in small groups. While this information was not asked, researchers were careful to interview only one member per group at residences or in shelters. It is believed that most respondents acted in small groups on 27 April, but exact numbers cannot be determined.

5. Conclusions

The 27 April 2011 EF4 Tuscaloosa, Alabama, tornado provided a unique opportunity to study how a violent tornado might change the shelter plans of residents for a future tornado event. Previous research has investigated where people seek shelter, how they respond to warnings, and the reasons for tornado fatalities. No previous research is believed to have exclusively focused on shelter-seeking plans before a tornado and stated changes to plans after a tornado.

An individual's shelter plan before 27 April was directly compared to changes in the individual's plan after 27 April. Only 99 (47%) of Tuscaloosa residents had shelter plans in place prior to 27 April, but 131 (62%) intend to change their plans for the future. Four groups were established to facilitate the reporting of results. Most residents did not know what their future plans were going to be at the time they were interviewed. Of those that responded, the majority of residents in the (yes, no) group plan to seek shelter within interior rooms in their homes in the future. The majority of residents in the (yes, yes) group plan either to purchase a storm shelter or to apply for a FEMA mitigation grant to construct a residential safe room. Most residents in the (no, yes) group plan to attempt to drive out of the path of an approaching tornado in the future. There were also many residents in this group who plan to apply for FEMA mitigation grants for residential storm shelters. The (no, no) group largely displayed fatalistic attitudes; however, a few respondents in this group mentioned that they might look at storm shelters or safe rooms after post-interview conversations with researchers.

Logistic regression results using social and demographic variables indicated that the strongest predictors of having a shelter plan before 27 April were age and education. Other variables were mostly insignificant. The strongest predictors of a future change in shelter plan were being Hispanic/Latino while having one's home destroyed were nearly significant.

In this research, a combination of qualitative and statistical methods was used to track individual changes to shelter-seeking plans and to make predictions about who might change their plans in the future. Using a combination of methods, it is hoped that NWS personnel, broadcast meteorologists, emergency managers, and city planners may use this research to make decisions about strategies to target particular groups of residents and potentially save lives in the future.

Local NWS personnel and broadcast meteorologists began discussing the potential for a historic tornado outbreak at least a week before 27 April 2011. Their efforts should be lauded, but many lessons were learned from that day. Local NWS field offices and broadcast meteorologists can potentially use the information in this research to consider making revisions to recommended shelter language used in warning messages on days when violent tornadoes are likely. Broadcast meteorologists are trusted by the public and were largely the preferred method by which respondents received warning information. They can adjust communication styles and methods for varying magnitudes of severe weather events to better emphasize the unique threat on violent tornado days. Emergency managers may initiate active discussions about shelter adequacy and ways to communicate shelter adequacy for typical and violent tornado days. City planners may pursue community shelter options and other creative ways to protect urban populations and increase preparedness. Finally, the most significant results in this research suggest that Hispanic/Latino residents may be particularly vulnerable to tornado hazards in the southern United States. This topic will be explored in future research.

Acknowledgments

This research was supported by National Science Foundation Grant 1138894. The authors thank Courtney Thompson, Laura Radford, Cory Rhodes, and Pilar Guinazu-Walker for their surveying, interviewing, and translating efforts. Special thanks to Dr. David Brommer for his efforts in developing the Zoomerang survey for use with the iPad. We would also like to thank four anonymous reviewers for their helpful comments.

REFERENCES

REFERENCES
Ashley
,
W. S.
,
2007
:
Spatial and temporal analysis of tornado fatalities in the United States: 1880–2005
.
Wea. Forecasting
,
22
,
1214
1228
.
Balluz
,
L.
,
L.
Schieve
,
T.
Holmes
,
S.
Kiezak
, and
J.
Malilay
,
2000
:
Predictors for people's response to a tornado warning: Arkansas, 1 March 1997
.
Disasters
,
24
,
71
77
.
Barnes
,
L. R.
,
E. C.
Gruntfest
,
M. H.
Hayden
,
D. M.
Schultz
, and
C.
Benight
,
2007a
:
Can close calls help define warning accuracy?
Bull. Amer. Meteor. Soc.
,
88
,
1529
1529
.
Barnes
,
L. R.
,
E. C.
Gruntfest
,
M. H.
Hayden
,
D. M.
Schultz
, and
C.
Benight
,
2007b
:
False alarms and close calls: A conceptual model of warning accuracy
.
Wea. Forecasting
,
22
,
1140
1147
;
Corrigendum, 24, 1452–1454
.
Biddle
,
M. D.
,
2007
:
Warning Reception, Response, and Risk Behavior in the 3 May 1999 Oklahoma City Long-Track Violent Tornado. ProQuest Information and Learning Company, 151 pp
.
Blanchard-Boehm
,
R. D.
, and
J. M.
Cook
,
2004
:
Risk communication and public education in Edmonton, Alberta, Canada on the 10th anniversary of the “Black Friday” tornado
.
Int. Res. Geogr. Environ. Educ.
,
13
,
38
54
.
Brotzge
,
J.
, and
S.
Erickson
,
2009
:
NWS tornado warnings with zero or negative lead times
.
Wea. Forecasting
,
24
,
140
154
.
Brotzge
,
J.
,
S.
Erickson
, and
H.
Brooks
,
2011
:
A 5-yr climatology of tornado false alarms
.
Wea. Forecasting
,
26
,
534
544
.
Carter
,
A. O.
,
M. E.
Millson
, and
D. E.
Allen
,
1989
:
Epidemiologic study of deaths and injuries due to tornadoes
.
Amer. J. Epidemiol.
,
130
,
1209
1218
.
Chaney
,
P. L.
, and
G. S.
Weaver
,
2010
:
The vulnerability of mobile home residents in tornado disasters: The 2008 Super Tuesday tornado in Macon County, Tennessee
.
Wea. Climate Soc.
,
2
,
190
199
.
Cohen
,
D.
, and
R. E.
Nisbett
,
1998
:
Are there differences in fatalism between rural Southerners and Midwesterners?
J. Appl. Soc. Psychol.
,
28
,
2181
2195
.
Coleman
,
T. A.
,
K. R.
Knupp
,
J.
Spann
,
J. B.
Elliott
, and
B. E.
Peters
,
2011
:
The history (and future) of tornado warning dissemination in the United States
.
Bull. Amer. Meteor. Soc.
,
92
,
567
582
.
Daley
,
W. R.
,
S.
Brown
,
P.
Archer
,
E.
Kruger
,
F.
Jordan
,
D.
Batts
, and
S.
Mallonee
,
2005
:
Risk of tornado-related death and injury in Oklahoma, May 3, 1999
.
Amer. J. Epidemiol.
,
161
,
1144
1150
.
Dow
,
K.
, and
S. L.
Cutter
,
1998
:
Crying wolf: Repeat responses to hurricane evacuation orders
.
Coastal Manage.
,
26
,
237
252
.
FEMA
, cited
2012
:
Building science. [Available online at http://www.fema.gov/rebuild/buildingscience/.]
Glass
,
R. I.
,
R. B.
Craven
,
D. J.
Bregman
,
B. J.
Stoll
,
N.
Horowitz
,
P.
Kerndt
, and
J.
Winkle
,
1980
:
Injuries from the Wichita Falls tornado: Implications for prevention
.
Science
,
207
,
734
738
.
Hall
,
S. G.
, and
W.
Ashley
,
2008
:
Effects of urban sprawl on the vulnerability to a significant tornado impact in northeastern Illinois
.
Nat. Hazards Rev.
,
9
,
209
219
.
Hammer
,
B.
, and
T. W.
Schmidlin
,
2002
:
Response to warnings during the 3 May 1999 Oklahoma City tornado: Reasons and relative injury rates
.
Wea. Forecasting
,
17
,
577
581
.
Hoekstra
,
S.
,
K.
Klocknow
,
R.
Riley
,
J.
Brotzge
,
H.
Brooks
, and
S.
Erickson
,
2011
:
A preliminary look at the social perspective of warn-on-forecast: Preferred tornado warning lead time and the general public's perceptions of weather risks
.
Wea. Climate Soc.
,
3
,
128
140
.
League
,
C. E.
,
W.
Díaz
,
B.
Philips
,
E. J.
Bass
,
K.
Kloesel
,
E.
Gruntfest
, and
A.
Gessner
,
2010
:
Emergency manager decision-making and tornado warning communication
.
Meteor. Appl.
,
17
,
163
172
.
Lindell
,
M. K.
, and
R. W.
Perry
,
2004
:
Communicating Environmental Risk in Multiethnic Communities. Sage Publications, 262 pp
.
National Weather Service Birmingham
, cited
2011
:
Historic tornado outbreak—April 27, 2011. [Available online at http://www.srh.noaa.gov/bmx/?n=event_04272011.]
Perduzzi
,
P.
,
J.
Concato
,
E.
Kemper
,
T. R.
Holford
, and
A. R.
Feinstein
,
1996
:
A simulation study of the number of events per variable in logistic regression analysis
.
J. Clin. Epidemiol.
,
49
,
1373
1379
.
Schmidlin
,
T. W.
, and
P.
King
,
1995
:
Risk-factors for death in the 27 March 1994 Georgia and Alabama tornadoes
.
Disasters
,
19
,
170
177
.
Schmidlin
,
T. W.
,
P.
King
,
B. O.
Hammer
, and
Y.
Ono
,
1998
:
Behavior of vehicles during tornado winds
.
J. Safety Res.
,
29
,
181
186
.
Schmidlin
,
T. W.
,
B.
Hammer
,
P.
King
,
Y.
Ono
,
L. S.
Miller
, and
G.
Thumann
,
2002
:
Unsafe at any (wind) speed? Testing the stability of motor vehicles in severe winds
.
Bull. Amer. Meteor. Soc.
,
83
,
1821
1830
.
Schmidlin
,
T. W.
,
B. O.
hammer
,
Y.
Ono
, and
P. S.
King
,
2009
:
Tornado shelter-seeking behavior and tornado shelter options among mobile home residents in the United States
.
Nat. Hazards
,
48
,
191
201
.
Schultz
,
D. M.
,
E. C.
Gruntfest
,
M. H.
Hayden
,
C. C.
Benight
,
S.
Drobot
, and
L. R.
Barnes
,
2010
:
Decision making by Austin, Texas, residents in hypothetical tornado scenarios
.
Wea. Climate Soc.
,
2
,
249
254
.
Schumacher
,
R. S.
,
D. T.
Lindsey
,
A. B.
Schumacher
,
J.
Braun
,
S. D.
Miller
, and
J. L.
Demuth
,
2010
:
Multidisciplinary analysis of an unusual tornado: Meteorology, climatology, and the communication and interpretation of warnings
.
Wea. Forecasting
,
25
,
1412
1429
.
Senkbeil
,
J. C.
,
D. M.
Brommer
,
P. G.
Dixon
,
M. E.
Brown
, and
K.
Sherman-Morris
,
2010
:
The perceived landfall location of evacuees from Hurricane Gustav
.
Nat. Hazards
,
54
,
141
158
.
Sherman-Morris
,
K.
,
2010
:
Tornado warning dissemination and response at a university campus
.
Nat. Hazards
,
52
,
623
638
.
Silver
,
A.
, and
C.
Conrad
,
2010
:
Public perception of and response to severe weather warnings in Nova Scotia, Canada
.
Meteor. Appl.
,
17
,
173
179
.
Simmons
,
K. M.
, and
D.
Sutter
,
2009
:
False alarms, tornado warnings, and tornado casualties
.
Wea. Climate Soc.
,
1
,
38
53
.
Sims
,
J. H.
, and
D. D.
Bauman
,
1972
:
The tornado threat: Coping styles of the North and South
.
Science
,
176
,
1386
1392
.
Sorenson
,
J. H.
,
2000
:
Hazard warning systems: Review of 20 years of progress
.
Nat. Hazards Rev.
,
1
,
119
125
.
Storm Prediction Center
, cited
2011a
:
Annual U. S. killer tornado statistics. [Available online at http://www.spc.noaa.gov/climo/torn/fataltorn.html.]
Storm Prediction Center
, cited
2011b
:
Monthly and annual U. S. tornado summaries. [Available online at http://www.spc.noaa.gov/climo/online/monthly/newm.html.]
Sutter
,
D.
, and
K. M.
Simmons
,
2010
:
Tornado fatalities and mobile homes in the United States
.
Nat. Hazards
,
53
,
125
137
.
Tobin
,
G. A.
, and
B. E.
Montz
,
1997
:
Natural Hazards: Explanation and Integration. Guilford Press, 387 pp
.

Footnotes

This article is included in the Tornado Warning, Preparedness, and Impacts Special Collection.